Nitride concentration quantitative analysis method based on relative position matrix method and Raman spectrum

Abstract

The invention relates to a nitride concentration quantitative analysis method based on a relative position matrix method and a Raman spectrum, and belongs to the field of Raman spectrum quantitative analysis. The method comprises the following steps: collecting Raman spectrum data of a sample through a Raman spectrometer, and establishing a Raman spectrum data set; carrying out smoothing and baseline deduction on the Raman spectrum data set, and carrying out data enhancement and data normalization on the data set; performing two-dimensional conversion on the preprocessed data set by using a relative position matrix method; training a multi-modal fusion neural network model based on the original one-dimensional spectral data and the converted two-dimensional image data; and quantitatively detecting the concentrations of nitrate and nitrite in water by using the trained model. The Raman spectrum and the relative position matrix method are combined, quantitative detection of the nitride concentration in the water sample is achieved, and technical support is provided for large-scale application of the Raman spectrum in actual water sample detection.

Description

Technical Field

[0001] The present invention belongs to the field of quantitative analysis of Raman spectroscopy, and relates to a method for quantitatively analyzing the concentration of nitride based on the relative position matrix method and Raman spectroscopy. Background Art

[0002] Nitrates and nitrites are common pollutants in surface water and groundwater. Excessive nitrate concentration can lead to eutrophication of water bodies, promote the rapid reproduction of plankton, consume dissolved oxygen in water bodies, and damage water quality. Long-term human absorption of nitrites can lead to an increase in methemoglobin in the blood, posing a carcinogenic risk. Therefore, accurately monitoring the concentrations of nitrates and nitrites in water bodies is of great significance. Nitrates and nitrites coexist in water nitrides, and it is not easy to detect them simultaneously using traditional detection methods. Raman spectroscopy can simultaneously detect the characteristic peaks of multiple substances and has natural advantages in detecting nitrides. As an inelastic scattering spectrum, Raman spectroscopy can provide fingerprint information of the sample structure and has the advantages of being non-destructive, fast, highly sensitive, and requiring no sample pretreatment. Therefore, it is widely used in fields such as disease diagnosis, substance identification, and drug detection. However, in practical applications, the intensity of the Raman scattering signal is weak. Although multiple substances can be detected simultaneously, the interference of overlapping peaks and background in the mixture makes it difficult to directly achieve quantitative detection of substances through Raman signals.

[0003] Traditional machine learning algorithms are difficult to meet the high-precision requirements for quantitative analysis in the Raman spectra of mixtures. Therefore, it is necessary to combine deep learning methods to achieve quantitative detection of nitrides in water. In recent years, deep learning technology has achieved state-of-the-art results in multiple fields, including natural language processing, image recognition, speech recognition, signal analysis, etc. Deep learning has a stronger model representation ability compared to machine learning and is more advantageous in the quantitative analysis task of Raman spectra. However, the components in actual water samples are complex, and the feature extraction ability of the deep learning method based on one-dimensional Raman spectra is limited, making it difficult to extract the features of overlapping peaks and weak peaks in the Raman spectra of mixtures, which hinders the quantitative detection of nitrides in actual water samples.

[0004] The deep learning method based on two-dimensional images can fully extract the features of overlapping peaks and weak peaks in the Raman spectra of mixtures. However, the existing wavelet coefficient method needs to screen and match wavelet bases according to the spectra, resulting in limited generality. Half of the information in the image generated by the Gramian angular field method is repeated, resulting in limited utilization of the image. Therefore, using a more general and efficient Raman spectroscopy analysis scheme can effectively promote the large-scale application of Raman spectroscopy in the detection of actual water samples. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method for quantitative analysis of nitride concentration based on the relative position matrix method and Raman spectroscopy, so as to overcome the problems that it is difficult to extract the characteristics of overlapping peaks and weak peaks in the Raman spectra of mixtures in the prior art, and the universality and efficiency of the existing two-dimensional conversion methods are limited, and to realize the two-dimensional conversion of Raman spectra using the relative position matrix method, and use a multi-modal fusion neural network model for simultaneous quantitative analysis of nitrate and nitrite concentrations in water.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for quantitative analysis of nitride concentration based on the relative position matrix method and Raman spectroscopy, the method comprising:

[0008] Step 1, collect Raman spectrum data of a potassium nitrate and sodium nitrite mixed solution through an optical experimental platform to establish a Raman spectrum data set;

[0009] Step 2, perform curve smoothing and baseline correction on the collected Raman spectrum data set, and perform data augmentation on the data set to expand the number of data set samples and sample diversity;

[0010] Step 3, apply the relative position matrix method to the preprocessed Raman spectrum data set for dimension conversion, and convert the one-dimensional Raman spectrum into two-dimensional image data;

[0011] Step 4, based on the one-dimensional Raman spectrum and two-dimensional image data, train a multi-modal fusion neural network model to extract features of different dimensions;

[0012] Step 5, input the Raman spectrum of the water sample to be measured into the trained multi-modal fusion model, and simultaneously obtain the nitrate and nitrite concentrations in the water sample.

[0013] Further, in the above step 1, the concentration categories of the prepared potassium nitrate and sodium nitrite solutions are not less than 9 types, and the number of spectra in the data set is not less than 1000;

[0014] Further, in the above step 2, the airPLS algorithm is used for baseline deduction of the Raman spectrum, and data augmentation is performed by simulating the Raman peak shift and background noise that occur during the spectrum acquisition process. In this algorithm, first perform a random horizontal displacement of the original Raman spectrum within the wavenumber range of [-10 cm -1 , 10 cm -1 , and then superimpose Gaussian white noise with a mean of 0 and a variance of 10 -5 to generate a new spectrum.

[0015] Further, in the third step, the Raman spectrum with a size of 1024×1 is cropped to 200×1, and the cropped spectrum is dimensionally transformed using the relative position matrix method. The size of the transformed image is 200×200, with a total of 40,000 pixel points.

[0016] Further, in the fourth step, the neural network model for multi-modal fusion consists of three parts. The first two parts extract one-dimensional and two-dimensional features respectively, and the third part is used for feature fusion.

[0017] The VGG module of the first part is used to extract two-dimensional image features, with VGG16 as the basic network architecture. The input layer has a size of 200×200, which is the same as the two-dimensional image size described in claim 4. The convolutional kernel size is 3×3, the pooling kernel size is 2×2, the activation function is LeakyReLU, there are two fully connected layers with 64 neurons each, and a Dropout layer with a dropout ratio of 0.2.

[0018] The LSTM module of the second part is used to extract one-dimensional spectral features. Its structure includes an input layer with a size of 200×1, which is the same as the one-dimensional spectrum size described in claim 4, an LSTM layer with 64 neurons, two fully connected layers with 64 neurons each, and a Dropout layer with a dropout ratio of 0.2.

[0019] The feature fusion module of the third part is used to fuse one-dimensional and two-dimensional features. Its structure includes a one-dimensional fusion layer that concatenates the features extracted by the LSTM layer in the second part with the result obtained from its fully connected layer; a two-dimensional fusion layer that concatenates the features extracted by the last convolutional layer of the VGG module in the second part with the result obtained from its fully connected layer; and a feature fusion layer that concatenates the results obtained from the one-dimensional fusion layer and the two-dimensional fusion layer.

[0020] The loss function for model training is MSE, and its calculation formula is:

[0021]

[0022] where N is the total number of samples, x i and y i represent the predicted result and the true result respectively.

[0023] Further, in the fifth step, the test set is input into the trained model for quantitative analysis of the concentrations of nitrate and nitrite, and the model performance is compared with the model that does not use the relative position matrix method.

[0024] The beneficial effects of the present invention are as follows:

[0025] The quantitative Raman spectroscopy analysis method of mixtures based on the relative position matrix method provided by the present invention converts one-dimensional Raman spectroscopy into a two-dimensional image by applying the relative position matrix method, which can fully exploit the characteristics of overlapping peaks and weak peaks in the Raman spectroscopy of mixtures, and has stronger generality and high utilization rate of image resources. The one-dimensional features and two-dimensional features are simultaneously extracted through a multi-modal fusion model, and finally, the simultaneous quantitative analysis of the concentrations of nitrate and nitrite in water is realized.

[0026] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following specification. Brief Description of the Drawings

[0027] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail preferably with reference to the accompanying drawings, where:

[0028] Figure 1 is the overall flowchart of the present invention.

[0029] Figure 2 is a schematic diagram of the average Raman spectrum of the dataset.

[0030] Figure 3 is a schematic diagram of the steps of data augmentation.

[0031] Figure 4 is a schematic diagram of the structure of the multi-modal fusion model.

[0032] Figure 5 is a two-dimensional image obtained by converting the one-dimensional Raman spectrum through the relative position matrix in the embodiment of the present invention. Detailed Embodiments

[0033] The following specific examples illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present invention schematically, and the following embodiments and the features in the embodiments can be combined with each other without conflict.

[0034] Among them, the attached drawings are only for illustrative purposes, showing only schematic diagrams rather than physical diagrams, and should not be construed as a limitation to the present invention; for better illustration of the embodiments of the present invention, some components in the attached drawings will be omitted, enlarged or reduced, which do not represent the dimensions of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the attached drawings may be omitted.

[0035] In the attached drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the attached drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the attached drawings are only for illustrative purposes and should not be construed as a limitation to the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0036] As Figures 1 to 5 shown, a quantitative analysis method of nitride Raman spectrum based on the relative position matrix method includes the following steps:

[0037] Step 1: Collect the Raman spectrum data of the potassium nitrate and sodium nitrite mixed solution through an optical experimental platform to establish a Raman spectrum data set.

[0038] The experimental reagents used are potassium nitrate standard solution (KNO 3 , 1000mg / L) and sodium nitrite solid (NaNO 2 powder). The solution is diluted with ultrapure water with a resistivity of 18.2MΩ. Nine groups of mixed solution samples with different concentrations of potassium nitrate: sodium nitrite of 100mg / L:100mg / L, 100mg / L:10mg / L, 70mg / L:70mg / L, 50mg / L:50mg / L, 30mg / L:30mg / L, 10mg / L:10mg / L, 10mg / L:1mg / L, 1mg / L:1mg / L, 1mg / L:0.1mg / L are prepared. The Raman spectrum data of the samples are collected using a He-Ne laser with a power of 10mw and a wavelength of 632.8nm and a Raman spectrometer with an integration time of 6s. Each concentration contains 140 spectral data, and a Raman spectrum data set with a total of 1260 data is established.

[0039] Step 2: Perform baseline correction and normalization on the collected Raman spectrum data set, and perform data enhancement on the data set to expand the sample quantity and sample diversity of the data set;

[0040] In the actual detection of Raman spectroscopy, there is often interference from the fluorescence background and the background noise of the instrument itself, which will interfere with the subsequent data analysis process. Therefore, it is necessary to perform baseline correction on the original Raman spectrum. The present invention uses the airPLS algorithm commonly used in Raman spectrum preprocessing as the baseline correction algorithm. This algorithm can balance the accuracy of the original data and the roughness of the fitting data, and adaptively calculate the weights in each iteration, so as to efficiently estimate and subtract the baseline.

[0041] To reduce the time cost of Raman spectrum acquisition, data augmentation is performed on the Raman spectrum dataset after baseline subtraction to expand the sample quantity and diversity. The specific steps are as follows:

[0042] 1. Classify the Raman spectrum dataset, and then expand it one by one according to each class.

[0043] 2. Select a sample sample in a certain class i , and then randomly select another sample sample in the same class j .

[0044] 3. Perform Raman peak shift simulation: Apply a horizontal displacement within the range of ±10 cm j to the sample sample -1 .

[0045] 4. Perform linear combination: Randomly generate a scaling factor α within the range of [0, 1]. Use α and 1 - α as weights to linearly weighted superimpose the sample sample i and the sample sample j to generate a new sample sample k , and the specific formula is as follows:

[0046] sample k = α × sample j + (1 - α) × sample j

[0047] 5. Perform background noise simulation: Superimpose Gaussian white noise with a mean of 0 and a variance of 10 k on the sample sample -5 to obtain new data.

[0048] The number of spectra for each concentration group is expanded from 140 to 200 after data augmentation, and the total number of spectra in the dataset is expanded from 1260 to 1800.

[0049] Finally, use min - max normalization to scale the intensity range of the Raman spectrum to the interval of [0, 1] to eliminate the influence of the order of magnitude on the subsequent model training. The normalization formula is:

[0050]

[0051] where x min and x max represent the minimum and maximum intensities in the original spectrum respectively, x represents the intensity of the original spectrum, and x norm is the value after normalizing the Raman spectrum intensity.

[0052] Step 3: Apply the relative position matrix method to the preprocessed Raman spectrum dataset for dimensional conversion, converting the one-dimensional Raman spectrum into two-dimensional image data.

[0053] Since the original Raman spectrum has 1024 data points, corresponding to a wavenumber range of 0 - 3755 cm -1 , and the characteristic peaks of potassium nitrate and sodium nitrite to be detected are located at 1050 cm -1 and 1330 cm -1 respectively. The original Raman spectrum contains a large number of wavenumber intervals that are irrelevant to the detection object. Therefore, a certain range of the original Raman spectrum is cropped. After cropping, 200 data points are retained, and its wavenumber range corresponds to 732 - 1455 cm -1 .

[0054] Apply the relative position matrix method to the one-dimensional spectrum after range cropping for two-dimensional conversion. By calculating the intensity difference between each wavelength and other wavelengths in the spectrum, a two-dimensional matrix containing relative information between different wavelengths can be obtained. Transform the spectrum with length n into a matrix of size n×n, where x n represents the intensity corresponding to each wavelength in the spectrum. The transformation equation is as follows:

[0055]

[0056] In this embodiment, the converted two-dimensional image is used as the dataset for training the neural network model, and the dataset is randomly divided. A total of 360 out of 1800 data are randomly selected as the test set. During the training process of the model, 5-fold cross-validation is adopted, that is, the remaining 1440 data are randomly divided into 5 groups, each group contains 288 spectral data. One group is selected as the validation set, and the remaining 4 groups are used as the training set to train the model. Repeat the above process 5 times to ensure that all 5 groups of data are used in the validation set. Finally, the average result of the 5 trainings is used as the final result of the model. This step can effectively avoid the accidental error caused by the dataset distribution and enhance the reliability of the model performance analysis.

[0057] Step 4: Based on the one-dimensional Raman spectrum and two-dimensional image data, train a multi-modal fusion neural network model to extract features of different dimensions. The specific steps are as follows:

[0058] The two-dimensional feature extraction module of the multi-modal fusion model uses the VGG network as the backbone. The input layer has a size of 200×200, which is the same as the size of the two-dimensional image generated in Step 3. The convolutional kernel size is 3×3, the pooling kernel size is 2×2, the activation function is LeakyReLU, and there are a total of 16 convolutional layers and 5 pooling layers. After the two-dimensional image is processed by the last pooling layer in the VGG module, a feature map with a size of 6×6×512 will be obtained, and then it is flattened to obtain a one-dimensional feature map F of 18432×1 2D Finally, after passing through two fully connected layers with 64 neurons and a Dropout layer with a dropout ratio of 0.2, the prediction result P is obtained 2D For subsequent feature concatenation.

[0059] The one-dimensional feature extraction module uses the LSTM network as the backbone. The input layer has a size of 200×1. After passing through an LSTM layer with 64 neurons, a one-dimensional feature map F with a size of 64×1 is obtained 1D After passing through two fully connected layers with 64 neurons and a Dropout layer with a dropout ratio of 0.2, the prediction result P is obtained 1D

[0060] The feature fusion module adopts the form of hybrid fusion. First, the feature map F of the last layer of the first two modules and the prediction result P of their respective fully connected layers are concatenated to obtain C 1D and C 2D Then, C 1D and C 2D are concatenated to obtain the final fused feature C Fusion . Finally, the fused feature C Fusion is input into two fully connected layers with 1 neuron each to obtain the concentration prediction results of nitrate and nitrite by the multi-modal fusion model

[0061] The activation function of the last fully connected layer of the multi-modal fusion model is Sigmoid, and the calculation formula of this function is:

[0062] Sigmoid(x) = (1 + e -x ) -1

[0063] This function can map the output of the model to the interval from 0 to 1, which is the same as the interval after maximum-minimum normalization in data preprocessing, facilitating the subsequent inverse normalization of the output results of the model to obtain specific substance concentration data

[0064] The loss function used in the model training process is MSE, and the calculation formula is:

[0065]

[0066] where N is the total number of samples, and x i and y i respectively represent the predicted result and the true result of the model for this sample. During the training process, the model searches for the optimal parameter weights by reducing the error between the predicted result and the true result, thereby improving the model accuracy.

[0067] Step Five: Input the Raman spectrum of the water sample to be measured into the trained multi-modal fusion model, and simultaneously obtain the nitrate and nitrite concentrations in the water sample. In this embodiment, the coefficient of determination R 2 is also selected as the evaluation index of the model, and its calculation formula is:

[0068]

[0069] where x i and y i respectively represent the predicted result and the true result, N is the total number of samples, and respectively represent the average values of the predicted result and the true result. The closer R 2 is to 1, the better the fitting degree and prediction performance of the model.

[0070] Table 1

[0071]

[0072] Table 1 shows the concentration prediction results of potassium nitrate and sodium nitrite by different methods. It can be seen from Table 1 that in this embodiment, for the multi-modal fusion model (MMF) combined with the Relative Position Matrix (RPM), the MSE and R 2 of the nitrate concentration prediction result on the validation set are 0.0078 and 0.9476, and the MSE and R 2 of the nitrite concentration prediction result are 0.0028 and 0.9753. The average MSE and R 2 are 0.0053 and 0.9615. Among them, the R 2 of MMF-RPM is improved by 0.0328 and 0.0336 respectively compared with CNN-1D and VGG-1D using one-dimensional Raman spectroscopy. And for CNN-RPM using the relative position matrix method, its R 2 is also improved by 0.0139, which proves the effectiveness of the present invention.

[0073] The present invention firstly creates a quantitative analysis method for Raman spectra of mixtures based on the relative position matrix method, that is, converting one-dimensional Raman spectra into two-dimensional images to fully exploit the characteristics of overlapping peaks and weak peaks in the Raman spectra of mixtures; establishing a multi-modal fusion neural network model to extract one-dimensional Raman spectral features and two-dimensional image features simultaneously, which can provide more comprehensive features compared with a single model and make up for the differences in a single modality; being able to effectively achieve the quantitative analysis of nitrates and nitrites in water and is expected to provide method support for the large-scale application of Raman spectroscopy in actual detection.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the present technical solution, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for quantitative analysis of nitride concentration based on relative position matrix method and Raman spectroscopy, characterized in that: The method includes: Step 1: Collect Raman spectral data of a mixed solution of potassium nitrate and sodium nitrite through an optical experimental platform to establish a Raman spectral data set; Step 2: perform curve smoothing and baseline correction on the collected Raman spectral data set, and perform data enhancement on the data set to expand the number and diversity of data set samples; Step 3, applying the relative position matrix method to the preprocessed Raman spectrum data set to perform dimensional conversion, converting the one-dimensional Raman spectrum into two-dimensional image data; Step 4: Based on the one-dimensional Raman spectrum and two-dimensional image data, a multimodal fusion neural network model is trained to extract features of different dimensions; Step 5: Input the Raman spectrum of the water sample to be tested into the trained multimodal fusion model, and obtain the nitrate and nitrite concentrations in the water sample.

2. The method for quantitative analysis of nitride concentration based on relative position matrix method and Raman spectroscopy according to claim 1, characterized in that: In the step 1, the concentration categories of the prepared potassium nitrate and sodium nitrite solutions are no less than 9, and the number of spectra in the data set is no less than 1000.

3. The method for quantitative analysis of nitride concentration based on relative position matrix method and Raman spectroscopy according to claim 1, characterized in that: In step 2, the airPLS algorithm is used to perform baseline subtraction on the Raman spectrum, and data enhancement is performed by simulating the Raman peak shift and background noise that occur during spectrum acquisition. In this algorithm, the original Raman spectrum is firstly [-10cm -1 ,10cm -1 ] random horizontal displacements within the wave number range, and then superimposed with a mean of 0 and a variance of 10 -5 The new spectrum is generated by Gaussian white noise.

4. The method for quantitative analysis of nitride concentration based on relative position matrix method and Raman spectroscopy according to claim 1, characterized in that: In the step three, the Raman spectrum with a size of 1024×1 is cropped to 200×1, and the cropped spectrum is dimensionalized using the relative position matrix method. The converted image size is 200×200, with a total of 40,000 pixels.

5. The method for quantitative analysis of nitride concentration based on relative position matrix method and Raman spectroscopy according to claim 1, characterized in that: In the step 4, constructing a multimodal fusion neural network model includes three parts, the first two of which extract one-dimensional and two-dimensional features respectively, and the third one is used for feature fusion; The VGG module of the first part is used to extract two-dimensional image features, with VGG16 as the basic network architecture, wherein the input layer size is 200×200, which is consistent with the two-dimensional image size described in claim 4, the convolution kernel size is 3×3, the pooling kernel size is 2×2, the activation function is LeakyReLU, two layers of fully connected layers with 64 neurons and a Dropout layer with a discard ratio of 0.2; The LSTM module of the second part is used to extract one-dimensional spectral features, and its structure includes an input layer with a size of 200×1, which is consistent with the one-dimensional spectrum size described in claim 4, an LSTM layer with 64 neurons, two fully connected layers with 64 neurons, and a Dropout layer with a discard ratio of 0.2; The feature fusion module in the third part is used to fuse one-dimensional and two-dimensional features. Its structure includes a one-dimensional fusion layer, which concatenates the features extracted by the LSTM layer in the second part with the results obtained by its fully connected layer; The two-dimensional fusion layer combines the features extracted by the last convolutional layer of the VGG module in the second part with the results obtained by its fully connected layer; The feature fusion layer combines the results of the one-dimensional fusion layer and the two-dimensional fusion layer; The loss function of model training is MSE, and the calculation formula is: Where N is the total number of samples, x i and i Represent the predicted results and the true results respectively.

6. The method for quantitative analysis of nitride concentration based on relative position matrix method and Raman spectroscopy according to claim 1, characterized in that: In the step 5, the test set is input into the trained model to perform quantitative analysis of nitrate and nitrite concentrations, and the model performance is compared with the model that does not use the relative position matrix method.

Images